Computer Olympiad 2017 Draughts

[Catherine]

Hi all,

Before all,
CONGRATULATIONS to Fabien and the others programmers for this wonderfull tounament.
Fabien said that all the engines had been improved, it's true, with this level so closer.

We don't know for the moment the improvement brought to Dragon, Michel can say something about the New Dragon?

Also, i thing, and i hope that i am not alone in this case, that the tournament would be more exciting with the participation of 3 others great programs : Kingsrow, Damage and Damy.

Of course, a program it's like a paint, it will never be OK (finished).

Catherine.

[BertTuyt]

Catherine, I certainly will participate in the future. For this year I have a sabbatical and focus on 8x8 Breakthrough Draughts (BT).
Maybe hereafter (when I strongly solved it), I will focus on 10x10 BT.

The only question I have for myself, how close programs like Scan (and others in the same league) are already to perfection.
Off course not 100% perfect, but that basically the difference between the top programs becomes so small that one need 1000 and more games to prove the difference (like checkers).
Like we have shared in this forum the game in itself might have a design flaw (the strength of the king), which leads to too many draws.
And also based upon this, computer draughts will not extremely surpass human performance based upon this draw margin.

I thank Fabien for providing alternatives like killer Scan and BT Scan.
And maybe we should revert one day (within our computer community) to one of these 2 alternatives.
I guess killer Draughts might be the best option, as I tend to believe that 10x10 BT Draughts (although perhaps even not weakly solvable), might be extremely strong when uses BT Databases and MidGame Databases.

Bert

[TAILLE]

Catherine, I certainly will participate in the future. For this year I have a sabbatical and focus on 8x8 Breakthrough Draughts (BT).
Maybe hereafter (when I strongly solved it), I will focus on 10x10 BT.

The only question I have for myself, how close programs like Scan (and others in the same league) are already to perfection.
Off course not 100% perfect, but that basically the difference between the top programs becomes so small that one need 1000 and more games to prove the difference (like checkers).
Like we have shared in this forum the game in itself might have a design flaw (the strength of the king), which leads to too many draws.
And also based upon this, computer draughts will not extremely surpass human performance based upon this draw margin.

I thank Fabien for providing alternatives like killer Scan and BT Scan.
And maybe we should revert one day (within our computer community) to one of these 2 alternatives.
I guess killer Draughts might be the best option, as I tend to believe that 10x10 BT Draughts (although perhaps even not weakly solvable), might be extremely strong when uses BT Databases and MidGame Databases.

Bert

Hi,

A agree with Bert, international 10x10 draughts is really a drawing game and as such it is not very motivating to continue to work on it is it?
What are the rules of Breakthrough Draughts ? Is it simply the international rules knowing that the game is won for the side reaching the king row ? That mean that a game cannot be a draw which seems very interresting!
Do you know which side has the advantage?

[BertTuyt]

Gerard, yep you are right, the first who reaches the other side wins.
So no draws.

The rules are (besides this one) the same.
Breakthrough Draughts on a 8x8 board is weakly and strongly solvable.
So I might later revert to 10x10.

Bert

[TAILLE]

Gerard, yep you are right, the first who reaches the other side wins.
So no draws.

The rules are (besides this one) the same.
Breakthrough Draughts on a 8x8 board is weakly and strongly solvable.
So I might later revert to 10x10.

Bert

Thank you for your answer, that sounds like a good challenge. Do you have generated an egdb?

[BertTuyt]

I'm working now on 8x8 on my slower computer ( i7-940 4 core 2.9 GHz) it takes 40 minutes to generate all up to 10P DBs for 8x8.
DB Solving speed > 8 MPositions/sec.
On my faster 8-core (4 Ghz) machine I expect it will take 15 minutes.
The current DB generator scales to 12P DB (for which I need 64GByte Memory).

I will modify it in the next weeks to cover up to 24P DB.
I reprogrammed the engine so it plays 8x8 BT.
Between move 8 and 12 I mostly see a DB win/loss, where I only use 1 core based search, and 15 seconds search time.
To early to tell if it is a theoretical win for white or black.

I expect that with a 12-14P DB, and a 8-core parallel search I might already weakly solve it.
Based upon the current speed it will take around 4 weeks on my faster computer to generate the 24P DB, to solve it strongly.

Bert

[Catherine]

Catherine, I certainly will participate in the future. For this year I have a sabbatical and focus on 8x8 Breakthrough Draughts (BT).
Maybe hereafter (when I strongly solved it), I will focus on 10x10 BT.

The only question I have for myself, how close programs like Scan (and others in the same league) are already to perfection.
Off course not 100% perfect, but that basically the difference between the top programs becomes so small that one need 1000 and more games to prove the difference (like checkers).
Like we have shared in this forum the game in itself might have a design flaw (the strength of the king), which leads to too many draws.
And also based upon this, computer draughts will not extremely surpass human performance based upon this draw margin.

I thank Fabien for providing alternatives like killer Scan and BT Scan.
And maybe we should revert one day (within our computer community) to one of these 2 alternatives.
I guess killer Draughts might be the best option, as I tend to believe that 10x10 BT Draughts (although perhaps even not weakly solvable), might be extremely strong when uses BT Databases and MidGame Databases.

Bert

Hi,

A agree with Bert, international 10x10 draughts is really a drawing game and as such it is not very motivating to continue to work on it is it?
What are the rules of Breakthrough Draughts ? Is it simply the international rules knowing that the game is won for the side reaching the king row ? That mean that a game cannot be a draw which seems very interresting!
Do you know which side has the advantage?

Hi,
I read your posts, and if i understood, the main reason that we not saw Damage and Damy is the fact that 10x10 draughts it's drawish !?
I don't know if it's like this with 8x8 or chess or any others board game, to be clear, my question is : that's it exist a draughts variation which can prove that the first or the second to play win ?

We saw that at a certain level, it's hard to win between to program, only to my point of view, if they uses the same kind of algorithm or eval function.
Why, only to take an exemple, slagzet it's crashed by Scan and Kingsrow itsn't at the same degree ?
It's seem that Kingsrow and many others engines integreted recently the pattern concept, the result it's that the level became closer.

I remember that, with the first dxp games against Scan and Kingsrow and the others engines, Scan was almost unbeattable. I mean that it don't lost a single game.

To conclude, i think that, the strenght of an engine it's decided by the eval function and the skill of the program. A game can be drawish.
Catherine

[TAILLE]

Catherine, I certainly will participate in the future. For this year I have a sabbatical and focus on 8x8 Breakthrough Draughts (BT).
Maybe hereafter (when I strongly solved it), I will focus on 10x10 BT.

The only question I have for myself, how close programs like Scan (and others in the same league) are already to perfection.
Off course not 100% perfect, but that basically the difference between the top programs becomes so small that one need 1000 and more games to prove the difference (like checkers).
Like we have shared in this forum the game in itself might have a design flaw (the strength of the king), which leads to too many draws.
And also based upon this, computer draughts will not extremely surpass human performance based upon this draw margin.

I thank Fabien for providing alternatives like killer Scan and BT Scan.
And maybe we should revert one day (within our computer community) to one of these 2 alternatives.
I guess killer Draughts might be the best option, as I tend to believe that 10x10 BT Draughts (although perhaps even not weakly solvable), might be extremely strong when uses BT Databases and MidGame Databases.

Bert

Hi,

A agree with Bert, international 10x10 draughts is really a drawing game and as such it is not very motivating to continue to work on it is it?
What are the rules of Breakthrough Draughts ? Is it simply the international rules knowing that the game is won for the side reaching the king row ? That mean that a game cannot be a draw which seems very interresting!
Do you know which side has the advantage?

Hi,
I read your posts, and if i understood, the main reason that we not saw Damage and Damy is the fact that 10x10 draughts it's drawish !?
I don't know if it's like this with 8x8 or chess or any others board game, to be clear, my question is : that's it exist a draughts variation which can prove that the first or the second to play win ?

We saw that at a certain level, it's hard to win between to program, only to my point of view, if they uses the same kind of algorithm or eval function.
Why, only to take an exemple, slagzet it's crashed by Scan and Kingsrow itsn't at the same degree ?
It's seem that Kingsrow and many others engines integreted recently the pattern concept, the result it's that the level became closer.

I remember that, with the first dxp games against Scan and Kingsrow and the others engines, Scan was almost unbeattable. I mean that it don't lost a single game.

To conclude, i think that, the strenght of an engine it's decided by the eval function and the skill of the program. A game can be drawish.
Catherine

Hi,

Yes Catherine the drawish character of draughts is an issue (at least for me). In order to take a significant advantage you need to have a far stronger eval function than your opponent which cannot be the case between the strongest programs. As I showed you, e.g. by the analysis I produced on the game Scan - Mobydam, the real difficulty of the game appeared when one side as a subsequent advantage but where it seemed we are exactly between the win and the draw. In such situation all the moves are difficult to choose and here we really can see the skillfullness of the programmers.

My feeling is that the breakthrough draughts is immediatly difficult at the first move of the game because it appears to me that we are exactly between the win and the loss. In these conditions the skillfull of the programmers is naturally highlighted.
I have just decided to try to program this new game!

[Catherine]

Hi,

A agree with Bert, international 10x10 draughts is really a drawing game and as such it is not very motivating to continue to work on it is it?
What are the rules of Breakthrough Draughts ? Is it simply the international rules knowing that the game is won for the side reaching the king row ? That mean that a game cannot be a draw which seems very interresting!
Do you know which side has the advantage?

Hi,
I read your posts, and if i understood, the main reason that we not saw Damage and Damy is the fact that 10x10 draughts it's drawish !?
I don't know if it's like this with 8x8 or chess or any others board game, to be clear, my question is : that's it exist a draughts variation which can prove that the first or the second to play win ?

We saw that at a certain level, it's hard to win between to program, only to my point of view, if they uses the same kind of algorithm or eval function.
Why, only to take an exemple, slagzet it's crashed by Scan and Kingsrow itsn't at the same degree ?
It's seem that Kingsrow and many others engines integreted recently the pattern concept, the result it's that the level became closer.

I remember that, with the first dxp games against Scan and Kingsrow and the others engines, Scan was almost unbeattable. I mean that it don't lost a single game.

To conclude, i think that, the strenght of an engine it's decided by the eval function and the skill of the program. A game can be drawish.
Catherine

Hi,

Yes Catherine the drawish character of draughts is an issue (at least for me). In order to take a significant advantage you need to have a far stronger eval function than your opponent which cannot be the case between the strongest programs. As I showed you, e.g. by the analysis I produced on the game Scan - Mobydam, the real difficulty of the game appeared when one side as a subsequent advantage but where it seemed we are exactly between the win and the draw. In such situation all the moves are difficult to choose and here we really can see the skillfullness of the programmers.

My feeling is that the breakthrough draughts is immediatly difficult at the first move of the game because it appears to me that we are exactly between the win and the loss. In these conditions the skillfull of the programmers is naturally highlighted.
I have just decided to try to program this new game!

Hi Taille,

Thank for answer, so with breakthought, that it's another concept, games can be less drawish. If it's the case, i think that it's a good idea to improve engines.

Catherine

[Rein Halbersma]

Thank for answer, so with breakthought, that it's another concept, games can be less drawish. If it's the case, i think that it's a good idea to improve engines.

Catherine

Not just less draws, zero draws

[BertTuyt]

Rein, I can confirm this, after several 8x8 BT games, still no draw

Bert

[Rein Halbersma]

Catherine, I certainly will participate in the future. For this year I have a sabbatical and focus on 8x8 Breakthrough Draughts (BT).
Maybe hereafter (when I strongly solved it), I will focus on 10x10 BT.

The only question I have for myself, how close programs like Scan (and others in the same league) are already to perfection.
Off course not 100% perfect, but that basically the difference between the top programs becomes so small that one need 1000 and more games to prove the difference (like checkers).
Like we have shared in this forum the game in itself might have a design flaw (the strength of the king), which leads to too many draws.
And also based upon this, computer draughts will not extremely surpass human performance based upon this draw margin.

I thank Fabien for providing alternatives like killer Scan and BT Scan.
And maybe we should revert one day (within our computer community) to one of these 2 alternatives.
I guess killer Draughts might be the best option, as I tend to believe that 10x10 BT Draughts (although perhaps even not weakly solvable), might be extremely strong when uses BT Databases and MidGame Databases.

Bert

Hi,

A agree with Bert, international 10x10 draughts is really a drawing game and as such it is not very motivating to continue to work on it is it?
What are the rules of Breakthrough Draughts ? Is it simply the international rules knowing that the game is won for the side reaching the king row ? That mean that a game cannot be a draw which seems very interresting!
Do you know which side has the advantage?

Martin Fierz (of the 8x8 checkers program Cake) calls this Breakthrough variation Kingscourt: http://checker-board.blogspot.nl/search ... Kingscourt He solved it for 8x8 checkers using only an opening book, without endgame databases!

[BertTuyt]

Rein, thanks for sharing.
Yes i have read the blog.
But think he weakly solved it, and I want to strongly solve it.
Next to that, but his attempt was years ago, it took him several months.
I will try to do it in 1 day, so generating of the DBs and then a parallel search.
Not sure it will work though

Bert

[Rein Halbersma]

Rein, thanks for sharing.
Yes i have read the blog.
But think he weakly solved it, and I want to strongly solve it.
Next to that, but his attempt was years ago, it took him several months.
I will try to do it in 1 day, so generating of the DBs and then a parallel search.
Not sure it will work though

Bert

Without any knowledge of the search tree, I doubt that you can weakly solve the game without at least the 16 pc dbs. The forward search in 8x8 checkers took 62 ply (to first king, no dbs). I would expect that you would need a much deeper search than 62 ply to forcibly reach the 8 vs 8 late middle game.

[BertTuyt]

Rein, as of today I see 8x8 BT database results between move 8 and 12, based upon a 1 core search, 15 seconds.
The plydepth I reach is 22-24 ply at that stage.
The DB is a 10P DB.

So calculating from move 1 that is 38 - 48 ply.
Assume that after every 4 moves (so 8 ply), 2 man disappear (due to capture).
So a 12P DB would give an advantage of about 8 ply, so than the search depth becomes 30-40 ply.
Knowing the hyper speed Joost Buijs realized, all does not seem impossible...

So I dont think I need to go to a 16P DB, but the proof of the pudding is in the eating.

Bert